Date: 2023-02-11 23:25
Sender: Mikael JaganI do hope that this year I'll be able to sit down and review our vignettes and help pages, and undertake the project of filling in the gaps. It should not be too hard, for instance, to draw some nice trees with tikz to more clearly display the output of showClass("Matrix").
In the mean time ...
The names of the non-virtual subclasses of denseMatrix are [dln](ge|tr|tp|sy|sp|po|pp)Matrix, p?Cholesky, and p?BunchKaufman.
The 'd' in [dln] and the (ge|tr|tp|sy|sp|po|pp) fragment are from LAPACK:
https://netlib.org/lapack/lug/node24.html
p?Cholesky and p?BunchKaufman are subclasses of dtrMatrix or dtpMatrix, reserved for triangular factorizations.
{The BunchKaufman factorization is only "triangular" in the sense that it can be specified by a vector of length n*(n+1)/2. {Well, by that logic, our denseLU class should be a subclass of dgeMatrix, but it isn't ... FIXME??}}
The names of the non-virtual subclasses of sparseMatrix are [dln][gts][CRT]Matrix, [dl]diMatrix, indMatrix, and pMatrix.
The [dln][gts][CRT] code is only LAPACK-like, as LAPACK does not know [CRT]. We are consistent with LAPACK in that we use the third letter to indicate the storage format (cf., dtr* versus dtp*).
The 'di' in [dl]diMatrix _is_ a LAPACK code, but I can't think of any LAPACK routine devoted to diagonal matrices.
indMatrix represents row index matrices and pMatrix is the subclass of indMatrix representing permutation matrices. Neither code is from LAPACK.
{ I'd really like to generalize indMatrix to also encompass column index matrices, by adding a slot named 'margin' (equal to 1L or 2L). It would simplify many of our methods, but also invalidate any serialized indMatrix out there ... sigh ... TODO }
The help page for class <name> is accessed by ?`<name>-class` or help("<name>-class"), the same for all documented S4 classes. Class definitions can be retrieved with getClassDef("<name>"). |
